Re: [Gnu-arch-users] conversion of bitkeeper archive to tla

[James Blackwell]

> On Sat, 2004-08-21 at 12:46, John Meinel wrote:
>> Well, speaking as someone who actively uses tla on cygwin, I can say it
>> might be there enough. It's significantly slower than arch on linux (at
>> least an order of magnitude.) but Lode Leroy's path compression version
>> does work in most situations.

Zenaan Harkness wrote:
> This leads me to think of the following concept - path compression with
> aliases of some sort (I haven't exactly started using tla yet - just
> launching into the tutorial as I write):

Unfortunately, the aliases that people have been talking about don't have
any impact on the arch on cygwin problems. The real problem is that the
cygwin guys haven't fix cygwin to use nt paths instead of win9x paths. 

Sure, the archive path shortening will help a bit, but rather than solve
the problem, it just moves the edge case around.

You'd have to go back in the list quite a bit (8-12 months), but there
was a long discussion about the path length problem on the list. I
believe the summary of the thread was that whats really called for is to
convince the cygwin guys to use longer NT paths rather than the 256 char
paths that are inherited from win9x.

> Have a storage layer "translation layer" that translates names (such as
> the lengthy names causing problems on Cygwin) into a filesystem name,
> and vice versa, as needed.

This can work reasonably well for the archives themselves. However, the
problem just resurfaces when somebody actually gets a branch from an
archive, and tries to hack around in a project tree.
 
> I would assume (but haven't used it enough to say, obviously) that paths
> might be able to be compressed in many ways (double dashes to single,
> split paths at logical 'directory' boundaries if they exist,...).

Yes. I haven't reviewed the code, but Lode Leroy has a patch that
handles the problem well from a conceptual basis. That one retains a
transparent layout (cat/branch/ver) while cutting back on the length
quite a bit.

> If there are circumstances where long names must be manually entered by
> the tla user, then perhaps the aliases would be useful there too.

Unfortunately, the aliases that people have been talking about don't have
any impact on the arch on cygwin problems. The real problem is that the
cygwin guys haven't fix cygwin to use nt paths instead of win9x paths. 

> Sorry if it's just rambling...
>

-- 
James Blackwell          Try something fun: For the next 24 hours, give
Smile more!              each person you meet a compliment!

GnuPG (ID 06357400) AAE4 8C76 58DA 5902 761D  247A 8A55 DA73 0635 7400